Bukwa has long been considered one of the oldest known fossil sites in East Africa, a radioisotopic date of ca 22 Ma having been obtained from a lava thought to overlie the sediments 〚7〛 and 〚10〛. However, the strata at Bukwa are probably considerably younger than 22 Ma, the mammalian fauna suggesting an age younger than Rusinga, which is 17.8 Ma 〚16〛. The few mammal fossils found by the Uganda Palaeontology Expedition (Bathyergoides neotertiarius, a large Palaeomeryx, Nguruwe kijivium, Kenyasus namaquensis, and Diamantohyus africanus) suggest an age of ca 17.5 Ma for the deposits. As such the fossil grassland at Bukwa is currently the oldest known in tropical Africa, and predates the previous record by about 1.5 Myr.

Bancroft 〚3〛, 〚4〛, 〚5〛 and 〚6〛 described fossil wood from various sites on the Ugandan side of Mount Elgon, while Chaney 〚12〛, 〚13〛 and 〚14〛 reported on fossil leaves from the Bugishu Sandstones, a prevolcanic deposit that crops out near Mbale, Uganda. Most of Bancroft’s material seems to have come from coarse agglomerates that erupted onto vegetation that was growing on the flanks of the volcano. The picture that emerged from these early studies was that Elgon was clothed in forest and woodland during the Early Miocene. However, agglomerates tend to over-represent macrobotanical remains and under-represent small plants, such as grasses, and the ages of the agglomerates are not well constrained, even if they are indeed most likely to be of Early Miocene age. They are younger than the Bukwa sites, since the latter immediately overlie basement complex gneisses.

Chaney’s material came from pre-volcanic sandstones and consists of a diverse leaf palaeoflora, comprising Caesalpiniaceae (Bauhinia, Berlinia, Cassia), Papilionaceae (Dalbergia), Rosaceae (Parinari), Combretaceae (Terminalia), Pittosporaceae (Pittosporum), and Dipterocarpaceae (Monotes and Marquesia). According to Bonnefille 〚9〛, such a flora occurs typically in Sudano-Zambezian woodlands that experience a seasonal climate with alternating dry and wet seasons, and thus it implies the presence of savannah, and thus grassland. The age of the Bugishu sandstone is unknown, but, being pre-volcanic, it is probably earlier than Bukwa, which consists of volcanic ashes overlying basement complex rocks.

Because of the sedimentological bias, it is interesting to examine plants preserved in finer grained sediments, such as those that occur at Bukwa. Hamilton 〚18〛 described fossil leaves and flowers that he assigned to Melianthiaceae (Bersema) and Sterculiaceae (Cola and Pterygota), on which basis he deduced that the Bukwa palaeoflora consisted of tropical lowland rainforest species, a conclusion supported by Bonnefille 〚9〛. Thus Bukwa has yielded an interesting ‘mélange’ of fossils, implying the existence of open and closed vegetation types at the time of deposition.

Most Early Miocene fossil sites in East Africa have yielded abundant evidence of hominoid primates 〚19〛, 〚21〛 and 〚22〛. Bukwa is exceptional in that very few primates have been recovered from it, the order being represented by only two isolated teeth. Previous literature 〚31〛 is somewhat misleading to the extent that some fragmentary suiform fossils (teeth of Diamantohyus africanus and Kenyasus namaquensis) were misidentified as Proconsul, which gave the false impression of the presence of these hominoids at the site. There are only two primate teeth from Bukwa: both are poorly preserved but have been identified as Limnopithecus legetet 〚19〛.

UMP 68–22 Bukwa II left lower m/1 Limnopithecus legetet UMP 68–26/27 Bukwa II upper left I1/ Limnopithecus legetet

It is possible that the rarity of primate fossils at Bukwa could be due to the fact that the depositional area was covered in grassland, with woodland and forest some distance away.

Terrestrial gastropods

Two field seasons of the Uganda Palaeontology Expedition at Bukwa in 1997 and 1998 resulted in the collection of numerous fossils, the most important of which were terrestrial gastropods. The 1967 collections from the site contained only 22 specimens, belonging to six taxa (Oreohomorus, Thapsia, Maizania, Tayloria, Achatina leakeyi and Burtoa nilotica), but previous identifications of these specimens contained several errors 〚31〛. A supposed Archachatina from Bukwa is more likely to represent Achatina leakeyi. The new collections comprise 136 additional specimens belonging to 15 genera (Table 1). From the enlarged assemblage, it is now possible to reconstruct the palaeoenvironment of Bukwa more satisfactorily than was hitherto possible.

The Bukwa gastropod fauna suggests that there was undoubtedly forest in the region, as shown by the presence of Ptychotrema, Tayloria, Trochozonites, and large Subulinidae (Oreohomorus and Subuliniscus) 〚23〛 and 〚28〛, but some of the forest may well have been rather open, or dry forest, because of the presence of Burtoa nilotica, Pseudoglessula and Maizania. Of particular interest is the discovery of a grassland type of Trochonanina preserved in the base of a clump of fossilised grass, similar to extant species that hide in the base of such clumps where they are protected from desiccation during dry seasons or from being burnt during grass fires. The presence of Helicarion and Chlamydarion also suggests relatively open conditions at Bukwa at the time of deposition.

Even if, at first glance, the snails that indicate open country at Bukwa appear to occur in relatively low numbers and are thus numerically subordinate to forest adapted lineages, two things need to be remembered. Firstly, large, robust shells of achatinids and subulinids tend to preserve better than small thin shells of urocyclids and helicarionids. This taphonomic bias can lead to under-representation of grassland species at the expense of woodland to forest types. Secondly, the population density and taxonomic diversity of snails in grasslands tends to be considerably lower than it is in forests and woodlands, and this can give a false impression about the relative importance of forest versus grassland species at a site. The combination of taphonomic and biological biases could thus yield the erroneous view that the sediments at Bukwa accumulated in a forested setting. The fossil grasses and the palaeosols indicate otherwise, and the few grassland types of snails support this conclusion.

Fossil grass occurs abundantly at Bukwa I. There are several fine-grained volcanic ash horizons with clumps of grass fossilised as imprints of leaves and three-dimensional stems in situ in their positions of growth with the root systems preserved in the underlying palaeosols. This indicates that there were grasslands at the site at the time of deposition. The closest analogy in terms of diameter and straightness of stems and size of clumps that we have found for the Bukwa grass fossils is the long grass plains of Karamoja, where grass grows in clumps rather than as a continuous cover. According to Vesey-Fitzgerald 〚29〛 present-day distribution of ‘clump’ or ‘bunch’ grasses (Panicum, Hyparrhenia, or Andropogon) is dependent more upon rainfall pattern and intensity than on fire. This type of grassland as a rule occurs where precipitation is between 500–1500 mm (20–60 inches) per year and where there is a marked annual dry season. Grass stems from Bukwa are straight and range in diameter from 1.6–3.2 mm. Further study of microscopic details of the leaves and stems is required for a definitive identification, but the gross morphology of the stems and leaves, as well as the clumped growth habitus, is sufficient to indicate that it could belong to one or other of the genera listed above, or to a species with similar growth habitus.

Near watercourses in grassy plains dominated by clump grasses, there is usually riparian forest, and this may well have been what Bukwa looked like at the time of deposition. The fossil fruits and other macrobotanical remains found by the UPE and by previous researchers accord with this scenario 〚18〛, 〚20〛, 〚30〛 and 〚31〛. It should be noted that the grass fossils at Bukwa are preserved where they grew, whereas the dicotyledon leaves, flowers, fruits and branches were transported unknown distances before deposition. Thus, when reconstructing the palaeoenvironment at Bukwa, greater significance should be given to the grass fossils than to the other plant remains. Fossil wood, fruit and seeds that occur in the Bukwa succession 〚18〛 indicate that there was lowland tropical forest in the vicinity. This environmental reconstruction accords with the evidence yielded by the gastropod fauna.

The fauna from the Bukwa I site confirms its subaerial affinities, with not a single aquatic element known in the assemblage. In contrast, Bukwa II, which is a paludal clay deposit of restricted areal extent, has yielded abundant remains of crabs and ostracods, but few mammals (both the hominoid specimens came from Bukwa II) and none of terrestrial gastropods. The limited areal extent of Bukwa II suggests that it was a small pond within the Bukwa depositional environment, most of which was sub-aerial.

As far as we know, Bukwa provides the earliest (ca 17.5 Ma) clear evidence of a grassland ecosystem in Eastern Africa, the older Bugishu ‘occurrence’ being implied from the dicotyledon flora, with no direct evidence of grasses at the site. The next oldest evidence is from Nyakach (ca 15 Ma) and Fort Ternan (12.5–13 Ma), Kenya.

Fossilised grass in fine-grained volcanic ash horizons at Bukwa (Early Miocene, Uganda), complete with the root systems in palaeosols that underlie each ash, provide the earliest good evidence for the existence of grassland ecosystems in tropical Africa. Previously, the earliest such ecosystem was recorded at Nyakach, (ca 15 Ma) in western Kenya 〚32〛.

Study of the fossil terrestrial gastropods from Bukwa supports the conclusion that the sediments accumulated in a grassland environment, but suggests that, in addition, there was woodland or forest in the vicinity at the time of deposition, a suggestion comforted by the macrobotanical remains found there 〚18〛.

For several reasons, the discovery of a grassland ecosystem in Early Miocene strata at Bukwa, Uganda, is important for understanding the evolution of vegetation types in the African tropics. Firstly, it effectively refutes the idea that tropical forest extended in an unbroken belt from the Atlantic to the Indian Ocean shores of Africa during the Early Miocene 〚1〛. It reveals that vegetation mosaics have been a feature of tropical Africa for at least 17.5 Ma 〚2〛. Secondly, the mammal fauna from Bukwa, with its grassland ecosystem, differs in several respects from Early Miocene faunas found in more wooded to forested locales, including Songhor, Koru, Rusinga and Napak, in which hominoid primates are a relatively common element 〚19〛, 〚21〛 and 〚22〛. Bukwa is unusual in having yielded very few primate fossils, but such a paucity of hominoids is more understandable when the grassland aspect of the palaeoenvironment is taken into account.

The presence of a grassland ecosystem at Bukwa pre-dates by about 10 Myr the origins of the family Hominidae, as well as the onset of the trend among African large mammals, such as Elephantidae and Suidae, to develop hypsodont cheek teeth endowed with cementum. Small mammals, such as lagomorphs (Austrolagomys), macroscelidids (Myohyrax), and some rodents (Diamantomys, Bathyergoides, Megapedetes) developed hypsodonty and cementodonty at a much earlier geological epoch, the earliest records of these taxa pre-dating the Bukwa deposits by up to 2 million years. It is thus not beyond the realms of possibility that evidence for the existence of grassland ecosystems even earlier than Bukwa will be found in East Africa when appropriate searches are made.